Pronab Kumar Paul | Representation Learning & Medical AI

About

I am an early-career researcher working at the intersection of computer vision, medical image analysis, and representation learning. My research focuses on understanding how AI models learn representations of medical images, why they fail under distribution shift, and how to develop trustworthy, robust, and clinically reliable medical AI systems.

I hold a B.Sc. in Information and Communication Engineering from the University of Rajshahi and currently work as a Research Assistant. My current work explores representation learning, multimodal medical imaging, and foundation models for healthcare applications.

I am seeking M.Sc. and Ph.D. opportunities in Computer Vision, Medical Image Analysis, and Representation Learning. I am also interested in related research opportunities in Healthcare AI, medical imaging, and intelligent IoT systems.

Highlights

Peer-reviewed publications in Springer CCIS and international engineering journals
Open-source research projects in medical image analysis, representation learning, and robust AI systems
Research on learned representation geometry, domain shift, multimodal medical imaging, and foundation models for healthcare

Current Research

I investigate the structure, geometry, and robustness of learned representations in medical imaging, studying how self-supervised and foundation models encode clinically relevant information. My work explores representation collapse, modality-specific learning dynamics, and failure under distribution shift, with the goal of developing reliable and interpretable AI systems for healthcare.

Representation Geometry Foundation Models Multimodal Medical Imaging Domain Shift Trustworthy AI Medical Image Segmentation

Selected Publications

Benchmarking the Robustness of Classical Keypoint Detectors and Descriptors across Real-World Imaging Scenarios

Paul, P.K., Ghosh, A., et al. · Preprint (Under Review), 2026

Preprint

Handling Imbalanced Datasets with Real-World Positive Samples for Dengue Prediction Using Machine and Deep Learning Models

Paul, P.K., Ghosh, A., et al. · Published in Springer CCIS, 2025

DOI

Design Optimization of a C/X-Band Metasurface-DGS Microstrip Antenna Using Machine Learning

Islam, M.B., Uddin, A.N.M.S., Bashir, S., Paul, P.K., et al. · Published in Physics Open, 2026

Scholar

Compact Patch Antenna for Wireless Sensor Networks

Ahmed, M.F., Bashir, S., Paul, P.K., et al. · Published in Applied Engineering & Technology, 2024

Scholar

Featured Research Projects

SSL Collapse & ACR Analysis

Investigating representation collapse in self-supervised medical foundation models using geometric metrics, including Average Cosine Rank (ACR), effective rank, and spectral analysis.

GitHub

Modality Collapse Framework

A reproducible framework for analyzing modality collapse and representation dynamics in multimodal medical imaging using layer-wise representation analysis and geometric similarity measures.

GitHub

Effective Rank & Synthetic Corruption

A research project investigating whether representation geometry, particularly effective rank, can predict performance degradation under synthetic image corruptions.

GitHub

TTA Failure Analysis

A systematic investigation of when test-time augmentation improves or degrades medical image classification, focusing on prediction stability, calibration, and transformation-induced failure modes.

GitHub

Medical Image Wavelet Analysis

A multiscale analysis framework using wavelet packet decomposition, spectral analysis, and fractal measures to characterize structural and frequency-domain properties of medical image modalities.

GitHub

Medical Image Spectral Analysis

A frequency and texture analysis framework using Fourier spectral analysis, Local Binary Patterns (LBP), and Gray-Level Co-occurrence Matrix (GLCM) descriptors to study modality-specific characteristics of medical images.

GitHub

Neuroimaging Dashboard

An interactive medical image segmentation application for visualizing neuroimaging data and generating tumor region segmentation masks. The current version is a lightweight demonstration, with ongoing work focused on improving robustness, speed, and efficiency.

GitHub

Research Interests

Computer Vision Medical Image Analysis Representation Learning Foundation Models Multimodal Medical Imaging Medical Image Segmentation Trustworthy AI Robust AI & Domain Shift

Education & Awards

B.Sc. in Information and Communication Engineering
University of Rajshahi
GPA: 3.82/4.00 · Academic Rank: 4th out of 65 (Top 6%)

Awards & Honors

General Merit Scholarship (2025-2026)
Awarded for academic excellence in B.Sc.
Dean's Award (2020-2025)
Awarded for sustained academic excellence at the University of Rajshahi.
Second Runner-Up, Learnathon 3.0
Developed a custom gRPC-based code execution engine.

Technical Skills

Programming

Python, C++

Deep Learning

PyTorch, TensorFlow, MONAI

Computer Vision

OpenCV

Machine Learning & Data Analysis

scikit-learn, NumPy, pandas

Research Computing

Google Colab, Jupyter Notebook, Matplotlib

Research Tools

Git, LaTeX, Overleaf

Contact

pronabpual77@gmail.com
s2010177116@ru.ac.bd

Department of Information and Communication Engineering,
University of Rajshahi